Purdue receives $10 million from National Science Foundation for Anvil supercomputer

WEST LAFAYETTE, Ind. — Pur­due Uni­ver­si­ty will soon be the home of Anvil, a power­ful new super­com­pu­ter that will pro­vi­de advan­ced com­pu­ting capa­bi­li­ties to sup­port a wide ran­ge of com­pu­ta­tio­nal and data-inten­si­ve rese­arch span­ning from tra­di­tio­nal high-per­for­mance com­pu­ting to modern arti­fi­cial intel­li­gence app­li­ca­ti­ons.

Anvil, which is fun­ded by a $10 mil­li­on award from the Natio­nal Sci­ence Foun­da­ti­on, will signi­fi­cant­ly incre­a­se the capa­ci­ty avail­ab­le to the NSF’s Extre­me Sci­ence and Engi­nee­ring Dis­co­very Envi­ron­ment (XSEDE), which ser­ves tens of thousands of rese­ar­chers across the U.S., and in which Pur­due has been a part­ner for the past nine years. Anvil will enter pro­duc­tion in 2021 and will ser­ve rese­ar­chers for five years. Addi­tio­nal fun­ding from the NSF will sup­port Anvil’s ope­ra­ti­ons and user sup­port.

 “Pur­due has a long histo­ry as a natio­nal lea­der in cam­pus super­com­pu­ting, and this award reflects that track record,” said The­re­sa May­er, Purdue’s exe­cu­ti­ve vice pre­si­dent for rese­arch and part­ners­hips. “High-per­for­mance com­pu­ting is cru­cial to dis­co­very in all domains today – sol­ving pro­blems in agri­cul­tu­re to sus­tainab­ly feed the world; in life sci­en­ces to under­stand and cure dise­a­se; and in engi­nee­ring to sup­port our natio­nal com­pe­ti­ti­ve­ness. We are plea­sed to have the oppor­tu­ni­ty to build upon our expe­ri­ence pro­vi­ding com­pu­ting resour­ces and exper­ti­se to cam­pus and bring it to bear enab­ling NSF rese­arch across the coun­try.”

Carol Song, the princi­pal inves­ti­ga­tor and pro­ject direc­tor for Anvil, and a seni­or rese­arch sci­en­tist for ITaP Rese­arch Com­pu­ting, says that Anvil was desi­gned to meet the natio­nal sci­ence and engi­nee­ring com­mu­nities’ needs for capa­ci­ty, acces­si­bi­li­ty and effec­ti­ve sup­port in today’s rapidly evol­ving com­pu­ta­tio­nal rese­arch land­s­cape.

We are exci­ted to bring a power­ful new com­pu­ting resour­ce to XSEDE rese­ar­chers and edu­ca­tors, espe­cial­ly tho­se in tra­di­tio­nal­ly under-ser­ved domains,” Song said.

By buil­ding Anvil along­side its com­mu­ni­ty clus­ter super­com­pu­ters, inclu­ding the 2020 “Bell” sys­tem being built for the Pur­due cam­pus, Pur­due will leverage its exis­ting cam­pus com­pu­ting infra­st­ruc­tu­re, such as mas­si­ve sto­rage sys­tems, high-speed net­wor­king, and its team of expert ITaP staff that has alrea­dy deploy­ed 14 lar­ge super­com­pu­ters sin­ce 2008. Purdue’s com­mu­ni­ty clus­ters ser­ve thousands of rese­ar­chers and stu­dents each year.

Pres­ton Smith, exe­cu­ti­ve direc­tor of rese­arch com­pu­ting and a co-PI on the pro­ject, says that Anvil will be opti­mi­zed for both the tra­di­tio­nal par­al­lel com­pu­ting that dri­ves rese­arch in are­as like flu­id dyna­mics and bio­in­for­ma­tics, and also for data sci­ence, arti­fi­cial intel­li­gence and machi­ne lear­ning app­li­ca­ti­ons.

The name ‘Anvil’ reflects the Pur­due Boi­ler­ma­kers’ strength and workm­an­li­ke focus on pro­du­cing results, and the Anvil super­com­pu­ter will enab­le important dis­co­ve­ries across many dif­fe­rent are­as of sci­ence and engi­nee­ring,” Smith said.

Anvil also will ser­ve as an expe­ri­en­ti­al lear­ning labo­ra­to­ry for stu­dents to gain real-world expe­ri­ence using com­pu­ting for their sci­ence, and for stu­dent interns to work with the Anvil team for con­struc­tion and ope­ra­ti­on. We will be trai­ning the rese­arch com­pu­ting prac­ti­tio­ners of the future,” he said.

Anvil will be built in part­ners­hip with Dell and AMD and will con­sist of 1,000 128-core AMD Epyc “Milan” pro­ces­sors and will deli­ver over 1 bil­li­on CPU core hours to XSEDE each year, with a peak per­for­mance of 5.3 peta­flops. Anvil’s nodes will be inter­con­nec­ted with 100 Gbps Mel­lanox HDR Infi­ni­Band. The super­com­pu­ter eco­sys­tem also will inclu­de 32 lar­ge memo­ry nodes, each with 1 TB of RAM, and 16 nodes each with four NVIDIA A100 Ten­sor Core GPUs pro­vi­ding 1.5 PF of sin­gle-pre­cisi­on per­for­mance to sup­port machi­ne lear­ning and arti­fi­cial intel­li­gence app­li­ca­ti­ons.

Rese­arch on Anvil will be able to leverage a diver­se set of sto­rage tech­no­lo­gies, ancho­red by a 10+ PB par­al­lel file­sys­tem, boos­ted with over 3 PB of flash disk. Sto­rage for acti­ve pro­jects and archi­val data will be pro­vi­ded by Purdue’s Rese­arch Data Depot and Fort­ress archi­ve. Novel work­flows will bene­fit from block and object sto­rage sys­tems also sup­por­ted by Anvil.

Anvil also will pro­vi­de a rich set of fea­tures that will broa­den access to advan­ced com­pu­ting,  inclu­ding inter­ac­ti­ve com­pu­ting and visua­liz­a­ti­on capa­bi­li­ties, and a ful­ly inte­gra­ted web-based Open OnDe­mand gate­way to Anvil’s soft­ware tools and com­pu­te nodes. A com­po­sable sub­sys­tem will enab­le cloud and con­tai­ner-based work­flows to run along­side the high-per­for­mance com­pu­ting sys­tem and will sup­port many modern sci­en­ti­fic app­li­ca­ti­ons, from gate­ways, data­ba­ses, high-through­put data ingesti­on pipe­lines, to com­plex cou­pled mode­ling work­flows. With seam­less pathways to Micro­soft Azu­re cloud, Anvil will help rese­ar­chers leverage both on-pre­mi­ses and com­mer­cial cloud com­pu­ting for advan­ced app­li­ca­ti­ons such as arti­fi­cial intel­li­gence mode­ling and infe­rence.

Amy Fried­lan­der, acting direc­tor of the Office of Advan­ced Cybe­r­in­fra­st­ruc­tu­re at NSF, said, “The­se awards repre­sent a suite of com­ple­men­ta­ry advan­ced com­pu­ta­tio­nal capa­bi­li­ties and ser­vices aimed to empower new fun­da­men­tal rese­arch in many fiel­ds. NSF’s long-stan­ding invest­ments in advan­ced and inno­va­ti­ve com­pu­ting respond to the rapid evo­lu­ti­on and expan­si­on of com­pu­ta­tio­nal- and data-inten­si­ve rese­arch being con­duc­ted across all of sci­ence and engi­nee­ring.”

The pro­ject is fun­ded under NSF award num­ber 2005632. Xiao Zhu, a com­pu­ta­tio­nal sci­en­tist and seni­or rese­arch sci­en­tist for rese­arch com­pu­ting, and Rajesh Kalya­nam, a data sci­en­tist and soft­ware engi­neer and rese­arch sci­en­tist for rese­arch com­pu­ting, are co-PIs on the pro­ject.

About Pur­due Uni­ver­si­ty

Pur­due Uni­ver­si­ty is a top public rese­arch insti­tu­ti­on deve­lo­ping prac­ti­cal solu­ti­ons to today’s toughest chal­len­ges. Ran­ked the No. 6 Most Inno­va­ti­ve Uni­ver­si­ty in the United Sta­tes by U.S. News & World Report, Pur­due deli­vers world-chan­ging rese­arch and out-of-this-world dis­co­very. Com­mit­ted to hands-on and online, real-world lear­ning, Pur­due offers a trans­for­ma­ti­ve edu­ca­ti­on to all. Com­mit­ted to afforda­bi­li­ty and acces­si­bi­li­ty, Pur­due has fro­zen tui­ti­on and most fees at 2012–13 levels, enab­ling more stu­dents than ever to gra­dua­te debt-free. See how Pur­due never stops in the per­sis­tent pur­su­it of the next giant leap at purdue.edu.

Wri­ter:  Adri­en­ne Mil­ler, sci­ence and tech­no­lo­gy wri­ter, Infor­ma­ti­on Tech­no­lo­gy at Pur­due (ITaP), 765–496-8204, mill2027@purdue.edu

Media con­ta­ct: Ste­ve Tal­ly, 765–494-9809, steve@purdue.edu, @sciencewriter